Twenty years ago, IBM’s computer “Deep Blue” defeated world chess champion Garry Kasparov, which sparked much discussion, attention, and speculation. People also began to challenge Go, which was seen as the “pinnacle of intelligence,” but due to the barriers faced by computers, breakthroughs had not been achieved. From March 9 to March 15, 2016, in just one week, Google’s AI robot AlphaGo played five matches against the world’s top Go player, Lee Sedol, and won by a score of 4:1, shocking the entire tech community. Artificial Intelligence (AI) and Artificial Neural Networks (ANN) became the focal points of discussion. The history of artificial intelligence can be traced back to the 1950s; it developed based on research from multiple disciplines including computer science, cybernetics, information theory, neuropsychology, philosophy, and linguistics, integrating new ideas, concepts, theories, and technologies into a comprehensive interdisciplinary frontier subject. AI is widely applied across various fields, including robotics, language recognition, military, and criminal investigation. The U.S. Defense Advanced Research Projects Agency (DARPA) gathered three teams at Harvard University, providing a total of $28 million in funding to find out why the human brain is better at learning than current artificial systems, aiming to develop more advanced AI. Dr. John Paulson of Harvard University believes that once we calculate the fundamental principles of how the brain learns, we will ultimately be able to design AI systems that can match or even surpass human intelligence. With the advancement of science and technology, the three major branches of AI technology—expert systems, artificial neural networks, and data deep mining—are playing increasingly significant roles in the medical field, drawing great attention.
1. Overview of Artificial Intelligence
In the summer of 1956, at a conference held at Dartmouth in the U.S., McCarthy first proposed the concept of “artificial intelligence” and regarded it as an emerging discipline. Since then, the concept of AI has been refined, but there has yet to be a consensus. Scholars such as Nilson at the AI research center of Stanford University define AI as a discipline that records, accumulates, reproduces, and applies knowledge by simulating human ways. The president of the Public University of Hakodate, Hidetaka Nakajima, defines AI as “machines or programs made artificially that possess intelligence, or disciplines that evaluate and study intelligence itself for the purpose of creating intelligence.” Professor Hiroshi Matsuo, a top AI expert and chairman of the Ethics Committee of the Japanese Society for Artificial Intelligence, believes that AI is “human-like intelligence created artificially,” meaning computers that have “discovery and perception functions,” capable of generating features from data and simulating related phenomena. These concepts reflect the scientific community’s basic understanding of AI. Over the past half-century, AI has made rapid advancements, achieving remarkable results, and is celebrated alongside atomic energy technology and space technology as one of the three major scientific breakthroughs of the 20th century. Some even refer to it as the “intelligence revolution”—a revolution that can lead to the emergence of an intelligent society.
2. The Prospects of AI and Expert Systems in Medicine
1. Development History and Application Examples of Medical Expert Systems
During the second AI boom, expert systems that fully utilize “knowledge” were continuously developed, refined, and utilized. An expert system is a program that introduces knowledge from a specific professional field and, through reasoning, can perform tasks as effectively as an expert in that field. Medical expert systems import a large amount of medical diagnostic knowledge into computers, simulating the clinical thought processes of medical experts, ultimately extracting and synthesizing valuable diagnostic clues from a knowledge base to provide treatment plans. One well-known medical expert system is the MYCIN system developed by Stanford University in the early 1970s, which diagnoses infectious disease patients and prescribes antibiotics. It contains 500 rules, and by sequentially answering its questions, the system can automatically determine the type of bacteria infecting the patient and prescribe the corresponding medication. Tests have shown that MYCIN’s diagnostic level for conditions like bacteremia, sepsis, lung infections, and brain infections has surpassed that of experts in the field. Recently, the Memorial Sloan-Kettering Cancer Center in the U.S. has been collaborating with IBM to introduce “Watson” technology to develop medical research applications, helping doctors select the best treatment plans for patients with special conditions. Researchers at the cancer center and IBM engineers are transmitting vast amounts of data related to conditions, treatment plans, and outcomes to Watson, using it to analyze the data and identify hidden patterns and correlations. Researchers hope that Watson can help doctors identify effective treatment plans, conduct clinical trials, publish trial results, and introduce these new treatment plans to doctors worldwide.
The development of medical expert systems in China began in the early 1980s, starting later than in developed countries but progressing rapidly. In 1978, Professor Guan Youbo from Beijing Traditional Chinese Medicine Hospital collaborated with experts in computer science to develop the “Guan Youbo Liver Disease Diagnosis Program,” the first application of a medical expert system in traditional Chinese medicine in China. In 1986, Professor Lin Rugaos’ student Lin Zishun assisted Fujian University of Traditional Chinese Medicine and the provincial computing center in inputting Lin Rugaos’ medical ideas into a computer, developing the “Lin Rugaos Bone Injury Diagnosis System,” which is among the advanced systems in the country. In 1992, the China Academy of Traditional Chinese Medicine and the Software Institute of the Chinese Academy of Sciences jointly developed the “Chinese Medicine Treatment Expert System.” In 1997, Shanghai United Traditional and Western Medicine Hospital collaborated with Yiyang Sheng Computer Company to develop the “Traditional Chinese Medicine Computer-Aided Diagnosis System,” which has now been integrated into many hospitals’ HIS systems. Since the 21st century, various medical expert systems have emerged, such as bone tumor auxiliary diagnosis expert systems, gastric cancer expert systems, periodontal disease diagnosis expert systems, cardiovascular drug treatment expert systems, and diagnosis systems based on spiral CT images of coronary artery calcification points.
2. Working Mechanism of Medical Expert Systems
It is generally believed that a traditional expert system consists of a knowledge base and an inference engine, hence expert systems are also referred to as knowledge and information-based systems. The expert knowledge stored in the knowledge base has fixed formal language expressions and data structure organization styles, primarily including three types: ① The most common is intuitive knowledge (experiential knowledge), often expressed as generative rules; when the conditions required by the rules are met, the system executes a certain action or draws a conclusion (the early version of MYCIN was like this); ② When the use of intuitive knowledge is difficult to solve complex problems, supportive knowledge is often used—medical theories that can guide medical practice, commonly represented by causal models; ③ Strategic knowledge, which can prioritize the use of certain rules when several rules are applicable simultaneously. The inference engine has two reasoning strategies: ① Forward reasoning, also known as data-driven reasoning, which derives new facts based on known facts by applying the rules whose conditions are satisfied, and then applies the relevant applicable rules of these new facts until an appropriate conclusion is reached; ② Backward reasoning, also known as hypothesis-driven reasoning, which first proposes a hypothetical conclusion and looks for rules whose conclusions match the hypothesis; the conditions required by these rules then become new hypotheses, and this process continues until all necessary hypotheses can be directly obtained from the user, thus confirming or denying some initial hypotheses. In complex clinical practice, many facts and conclusions do not have absolute certainty, and statistical reasoning or fuzzy reasoning is often required; the system’s reasoning is not deterministic but proposes the credibility of each conclusion, prioritizing those with higher credibility. For particularly complex and difficult problems, the system can also provide several possible conclusions and their credibility for medical staff to consider. This is significantly meaningful in specific clinical cases; for example, the new version of the MYCIN system contains such mechanisms.
3. The Prospects of AI and Artificial Neural Networks in Medicine
1. Background and Overview of Artificial Neural Networks
During the second AI boom, as long as enough knowledge was input into the computer, it could correspondingly complete many tasks, but its capabilities were limited to the scope of the input knowledge; to expand the computer’s practicality and its ability to handle exceptional cases, massive amounts of knowledge needed to be input, which could never be exhausted. Moreover, fundamentally, the symbols input often disconnect from their meanings, making it very difficult for computers to grasp “semantics.” However, under these closed conditions, a technology has steadily developed—machine learning, which allows AI programs to learn by themselves. Common principles of machine learning include nearest neighbor classification algorithms, naive Bayes algorithms, decision trees, and support vector machines. Among these, the most notable is the Artificial Neural Network (ANN). ANN emerged during the third AI boom as a highly integrated interdisciplinary frontier subject that combines brain science, information science, and computer science. It is an information processing system that abstracts and synthesizes the theories of biological neural networks in terms of structure and function by mimicking human neural circuits, representing an important branch of contemporary AI.
2. Advantages of Artificial Neural Networks
Compared to traditional symbolic processing methods, ANN has unique advantages: ① Distributed information storage; ANN presents and processes information through the connections and weights between neurons, making it highly stable and less affected by local network failures; ② Adaptability; the entire ANN can self-adjust based on the current environmental state and information characteristics, including learning, self-organization, generalization, and training. ANN continuously establishes new patterns that align with external changes through learning, systematically and efficiently connecting and distributing multiple neurons through self-organization. Generalization refers to ANN’s ability to respond most reasonably to entirely new information inputs through continuous training; ③ Parallelism; while processing information, each neuron in ANN cooperates, forming a network synergy while maintaining its independence and sharing output results with other neurons; ④ Associative memory function, which can accomplish complex nonlinear mappings, making it an ideal nonlinear estimator, and can learn adaptively, allowing the network to exhibit abstract thinking abilities and perform associative reasoning.
3. Application Examples of Artificial Neural Networks in Medicine
Due to its ability to effectively overcome the problem of “limited knowledge input” and its capabilities in learning, self-organization, generalization, and training, ANN has rapidly developed in the field of medical expert systems. In the intelligent recognition of medical images, videos, and audio, the “Artificial Retina” launched by Mitsubishi Electric’s LSI manufacturing center can accurately, efficiently, and rapidly process massive amounts of unstructured medical data. For medical diagnosis, a medical diagnostic expert system based on the PDP model developed by Professors Saito and Nakano from King Saud University in Saudi Arabia has achieved diagnostic accuracy far exceeding that of traditional medical expert systems, rivaling the most knowledgeable experts in related fields. The DP neural network myoelectric pulse identification program developed by Steven and others also achieves diagnostic accuracy far surpassing traditional machine recognition methods. In traditional Chinese medicine, the ANN-based diagnosis system can intelligently “differentiate syndromes” to a certain extent, proposing reasonable Chinese medical diagnoses after comprehensive analysis. Recently, IBM’s research team designed a user interface called “WatsonPaths” based on ANN machine learning principles. “WatsonPaths” is a human-computer interaction program that helps Watson learn how to diagnose and treat patients. With the assistance of “WatsonPaths,” medical staff can check whether the symptoms and inferences presented by Watson are reasonable, then input more information and insights into the Watson system. Meanwhile, the Memorial Sloan-Kettering Cancer Center, WellPoint, and IBM have developed interactive tumor treatment technologies based on Watson technology. This technology continuously helps oncologists gain the latest treatment information for patients through deep learning based on ANN technology. The database encompasses a wide range of information, including medical records, large libraries of medical literature, clinical guidelines, top doctors’ notes, and drug trial reports. Currently, the Watson system has incorporated 600,000 pages of medical reports, 42 medical journals, nearly 2 million pages of medical papers and clinical trial reports, as well as tens of thousands of medical records.
4. Future Development Suggestions for AI in Healthcare
The rapid development of AI technology in recent years has made the development and application of medical expert systems and artificial neural networks in the medical field a reality, achieving significant breakthroughs. However, currently, in China, the development trend and application scale of medical AI still have a considerable gap compared to Western developed countries, with generally low technical levels, most belonging to low-level development, and there is still much room for improvement in performance. Closer integration with clinical practice is also needed. AI is at the forefront of computer science, and its continuous development and application in the medical field require the joint efforts of experts in software and hardware, medical specialists, mathematicians, and others, necessitating cross-disciplinary collaboration. On one hand, more mature algorithms should be applied to help expert systems assist doctors in accurately and scientifically identifying effective treatment plans; on the other hand, research on ANN should continue to strengthen its capabilities for learning, self-organization, generalization, and training.
(This article was written by Dr. Kong Xiangyi from the Neurosurgery Department of Peking Union Medical College Hospital, and reviewed by Chief Wang Renzhi)
References
[1] Ensmenger N. Is chess the drosophila of artificial intelligence? A social history of an algorithm[J]. Soc Stud Sci,2012,42(1):5-30.
[2] Lawrence D R, Palacios-Gonzalez C, Harris J. Artificial Intelligence[J]. Camb Q Healthc Ethics,2016,25(2):250-261.
[3] de Grey A D. Artificial intelligence and medical research: time to aim higher?[J]. Rejuvenation Res,2016.
[4] Das S, Roy C S, Saha H. Accuracy enhancement in a fuzzy expert decision making system through appropriate determination of membership functions and its application in a medical diagnostic decision making system[J]. J Med Syst,2012,36(3):1607-1620.
[5] Furiasse N, Thomas J D. Automated Algorithmic Software in Echocardiography: Artificial Intelligence?[J]. J Am Coll Cardiol,2015,66(13):1467-1469.
[6] Jagarlapudi S A, Kishan K V. Database systems for knowledge-based discovery[J]. Methods Mol Biol,2009,575:159-172.
[7] Hua J, Hao Y, Jin P, et al. [Challenges to evidence-based medicine in an era of artificial intelligence][J]. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue,2015,27(9):709-711.
Neurosurgery News WeChat public subscription number can be pinned to the top!!! Open WeChat (version 6.3.16 and above), find the subscription number of Neurosurgery News, open it and click the profile picture in the upper right corner, enter the detailed information page of Neurosurgery News, and choose to pin the public account to the top. Then you will never miss important information!
“Neurosurgery News” may not be reproduced without permission. If you wish to reprint, please contact “Neurosurgery Assistant.” Please respect the copyright of “Neurosurgery News” when reprinting articles from various WeChat public accounts, websites, and clients. When permitted to reprint articles, please clearly indicate the source as “Neurosurgery News.” “Neurosurgery News” welcomes you to share in your circle of friends.
